Risk prediction models for incident heart failure: a systematic review and meta-analysis.

Area of Science:

• Cardiology
• Epidemiology
• Biostatistics

Background:

• Early identification of patients at risk of heart failure (HF) is crucial for preventative management.
• Existing HF incidence prediction models require further validation.
• This study systematically reviews the performance of HF risk prediction models in validation studies.

Purpose of the Study:

• To summarize the performance of risk prediction models for incident heart failure (HF) as observed in validation studies.
• To assess the discrimination and calibration of various HF prediction models.
• To identify promising models for future validation and impact studies.

Main Methods:

• Systematic search of Medline and Embase databases (2014-2025) in addition to previous reviews.
• Inclusion of derivation or validation studies for incident HF prediction models validated in at least one cohort.
• Meta-analyses for pooled discrimination and descriptive summaries for calibration; risk of bias assessed using PREDICT.
• Certainty of evidence assessed using Grading of Recommendations, Assessment, Development and Evaluation (GRADE).

Main Results:

• 76 studies representing 238 models were included; 82.9% of assessments showed high risk of bias.
• Among 64 models validated in at least one cohort, eight showed high discrimination and four showed moderate discrimination.
• The PREDICT, ARIC, and MESA models were identified as most promising.
• Externally validated models included machine learning (14), novel biomarkers (6), and social determinants of health (9).

Conclusions:

• The PREDICT, ARIC, and MESA risk scores demonstrate promising performance for predicting heart failure incidence.
• These models should be prioritized for further external validation.
• Progression to impact studies is recommended to assess their clinical utility in preventative management.